The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides the agent with high levels of autonomy, instructing it to act as the primary builder with terminal access. It mandates silent execution of commands like npm install, cdk synth, and cdk deploy --require-approval never without requiring specific user confirmation for each step.- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install external dependencies via npm and references the download of the AWS CLI installer and an S3 Tables catalog JAR from well-known and trusted vendor sources.- [COMMAND_EXECUTION]: The agent is expected to execute a variety of shell commands, including AWS CLI operations for stack verification, Glue job triggers, and Athena query executions.- [PRIVILEGE_ESCALATION]: The documentation in reference/gotchas.md includes instructions for the agent to upgrade the local AWS CLI using sudo installer, which involves gaining administrative privileges to modify system files.- [INDIRECT_PROMPT_INJECTION]: The skill possesses a broad data ingestion surface, processing files from S3, JDBC databases, and SAP OData services. While the skill includes data normalization and validation logic in reference/scripts.md, the processing of external data by an agent with terminal execution capabilities creates a potential vector for indirect prompt injection attacks.
Ingestion points: S3, JDBC, and SAP sources described in SKILL.md §1 and §3.
Boundary markers: The skill employs database namespaces ({prefix}_db) and Iceberg table properties to delimit data storage.
Capability inventory: Terminal access for cdk and npm, boto3 for Glue/Athena operations, and sudo for tool installation.
Sanitization: The skill provides complex regex and parsing scripts in reference/scripts.md to handle malformed strings, date formats, and numeric values common in ERP data.

data-platform-pipeline